| description abstract | Abstract. This study proposes an integrated approach for predicting the crack driving force in bimetallic welded pipes by combining the finite element method (FEM) and machine learning. The bimetallic pipes are connected by welding, often leading to flaws and misalignment, which present significant challenges to their integrity. In this work, a study was conducted on the factors affecting the crack tip opening displacement (CTOD) of geometrically mismatched composite pipes with surface cracks. A numerical framework for canoe-shaped surface cracks was developed. After verifying the accuracy of the method, a reasonable parameter space was designed, and axial tensile finite element simulations were performed under high-strain conditions. The linear coefficients and c2 of the linear segment of the CTOD-εg curve were used as output variables. Different machine learning models were employed for training, and the impact of each feature on crack response was analyzed based on permutation importance. The multilayer perceptron (MLP) model showed the highest prediction accuracy, and the crack depth ratio was consistently identified as the most influential feature. By combining the finite element analysis with machine learning models, the crack driving force for canoe-shaped surface cracks in oil and gas pipelines entering the plastic deformation stage can be predicted more accurately and quickly. This approach strikes a balance between accuracy and overly conservative tactics, providing reliable technical support for pipeline design and safety assessments. | |
